On January 15, 2026, TSMC closed out 2025 with the strongest quarter in its history. Fourth-quarter revenue reached NT$1.046 trillion (US$33.73 billion), up 20.5% year over year — the first time the company has crossed NT$1 trillion in a single quarter. Net income rose 35% to NT$505.74 billion (about US$16.05 billion), the eighth straight quarter of year-over-year profit growth and another record. Both figures beat the analyst estimates compiled by LSEG.
For anyone tracking the AI buildout, TSMC’s report is the hardest signal to fake: every data center GPU, accelerator, and AI-capable device eventually becomes an order for advanced-node wafers. Add the company’s 2026 capital expenditure guidance of US$52–56 billion, and the message is unambiguous — AI demand is showing up as signed purchase orders, not slideware.
First Quarter Above NT$1 Trillion
The quarter’s quality matched its size. Gross margin came in at 62.3%, beating the company’s own guidance range of 60.5% to 62.5% — pricing power that reflects how few foundries can produce leading-edge AI silicon at volume. JPMorgan said in a post-earnings note that TSMC will continue to ride US AI demand and extend its lead over rivals such as Samsung and Intel. CFO Wendell Huang told analysts that AI chip demand keeps the supply picture tight and that visibility keeps improving.
Guidance points the same direction: TSMC expects first-quarter 2026 revenue of US$34.6–35.8 billion, up about 4% sequentially and roughly 38% year over year at the midpoint — an unusually steep ramp for a seasonally soft quarter.
The 2-Nanometer Ramp and Advanced Node Mix
Volume production of 2-nanometer chips began in the fourth quarter and will ramp further through 2026. Chips at 7 nanometers and below — the tiers that carry AI processors — accounted for 77% of wafer revenue in Q4. For full-year 2025 the figure was 74%, up from 69% in 2024, a mix shift that explains both the record margins and the surging capex plan.
By platform, high-performance computing, which includes AI and 5G chips, contributed 55% of fourth-quarter sales, against 32% from smartphones. AI silicon, not handsets, is now the center of gravity of the world’s most advanced factory network.
The $52–56 Billion Capex Plan
TSMC guided 2026 capital expenditure to US$52–56 billion, up from the US$40.9 billion spent in 2025 — a jump of roughly a third at the midpoint. Counterpoint Research’s Jake Lai called 2026 a “breakout year” for AI server demand, while flagging the global memory shortage as the main near-term threat, since it could crimp consumer electronics supply.
The spending plan is effectively a forward order book for the AI economy: equipment vendors, construction, power, and packaging capacity all get pulled through it. When the foundry at the bottom of the stack commits at this scale, the bottleneck moves elsewhere — to memory, power, and advanced packaging, which is exactly where shortages are already appearing.
US Expansion vs. Gross Margin
Chairman C.C. Wei acknowledged the industry-wide memory shortage on the earnings call, saying the shortfall is most acute for mainstream Node B and embedded NVM capacity that goes into consumer devices, while premium products built around AI silicon are less exposed; some customers have responded to constrained memory supply by placing increased orders for AI-focused products. He also flagged tariffs as a risk to keep watching.
On the US buildout, TSMC has purchased additional land in Arizona, and Wei reiterated the vision of the site as a “gigafab” cluster, with significant expansion of overseas manufacturing capacity coming in 2026. The price is gross margin dilution from overseas fabs — about 2 percentage points in 2025, higher in 2026 — the cost of geographic diversification written directly into the income statement.
What It Means for the AI Supply Chain
Three reads for builders. First, advanced-node supply stays tight through 2026; if your roadmap depends on leading-edge silicon, lock capacity early. Second, the adjacent risk is memory — the shortage bites mainstream consumer memory parts rather than AI accelerators, but it reshapes BOM costs for edge AI devices. Third, TSMC’s capex is the most reliable public proxy for AI infrastructure demand, and a commitment of this size keeps the 2026 spending supercycle thesis from our opening-of-year outlook on track.
Sources
- TSMC extends record quarter, profit jumps 35% on robust AI chip demand — CNBC
- TSMC smashes forecasts with record profit, flags more US investment — Reuters
- Q4 2025 quarterly results — TSMC Investor Relations
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.